mercurial/changegroup.py

changegroup: move chunk extraction into a getchunks method of unbundle10

This code used to be in `writebundle` only. We needs to make it more broadly
available for bundle2. The "changegroup" bundle2 part has to retrieve the
binary content of changegroup stream. We moved the chunks retrieving code into
the `unbundle10` object directly and the `writebundle` code is now using that.

This split is useful for bundle2 purpose, we want to be able to easily stream
changegroup content in a part.

To keep thing simples, we kept compression out of the new methods. If it make
more sense in the future, compression may get included in this function too.

# changegroup.py - Mercurial changegroup manipulation functions
#
#  Copyright 2006 Matt Mackall <[e-mail]>
#
# This software may be used and distributed according to the terms of the
# GNU General Public License version 2 or any later version.

import weakref
from i18n import _
from node import nullrev, nullid, hex, short
import mdiff, util, dagutil
import struct, os, bz2, zlib, tempfile
import discovery, error, phases, branchmap

_BUNDLE10_DELTA_HEADER = "20s20s20s20s"

def readexactly(stream, n):
    '''read n bytes from stream.read and abort if less was available'''
    s = stream.read(n)
    if len(s) < n:
        raise util.Abort(_("stream ended unexpectedly"
                           " (got %d bytes, expected %d)")
                          % (len(s), n))
    return s

def getchunk(stream):
    """return the next chunk from stream as a string"""
    d = readexactly(stream, 4)
    l = struct.unpack(">l", d)[0]
    if l <= 4:
        if l:
            raise util.Abort(_("invalid chunk length %d") % l)
        return ""
    return readexactly(stream, l - 4)

def chunkheader(length):
    """return a changegroup chunk header (string)"""
    return struct.pack(">l", length + 4)

def closechunk():
    """return a changegroup chunk header (string) for a zero-length chunk"""
    return struct.pack(">l", 0)

class nocompress(object):
    def compress(self, x):
        return x
    def flush(self):
        return ""

bundletypes = {
    "": ("", nocompress), # only when using unbundle on ssh and old http servers
                          # since the unification ssh accepts a header but there
                          # is no capability signaling it.
    "HG10UN": ("HG10UN", nocompress),
    "HG10BZ": ("HG10", lambda: bz2.BZ2Compressor()),
    "HG10GZ": ("HG10GZ", lambda: zlib.compressobj()),
}

# hgweb uses this list to communicate its preferred type
bundlepriority = ['HG10GZ', 'HG10BZ', 'HG10UN']

def writebundle(cg, filename, bundletype, vfs=None):
    """Write a bundle file and return its filename.

    Existing files will not be overwritten.
    If no filename is specified, a temporary file is created.
    bz2 compression can be turned off.
    The bundle file will be deleted in case of errors.
    """

    fh = None
    cleanup = None
    try:
        if filename:
            if vfs:
                fh = vfs.open(filename, "wb")
            else:
                fh = open(filename, "wb")
        else:
            fd, filename = tempfile.mkstemp(prefix="hg-bundle-", suffix=".hg")
            fh = os.fdopen(fd, "wb")
        cleanup = filename

        header, compressor = bundletypes[bundletype]
        fh.write(header)
        z = compressor()

        # parse the changegroup data, otherwise we will block
        # in case of sshrepo because we don't know the end of the stream

        # an empty chunkgroup is the end of the changegroup
        # a changegroup has at least 2 chunkgroups (changelog and manifest).
        # after that, an empty chunkgroup is the end of the changegroup
        for chunk in cg.getchunks():
            fh.write(z.compress(chunk))
        fh.write(z.flush())
        cleanup = None
        return filename
    finally:
        if fh is not None:
            fh.close()
        if cleanup is not None:
            if filename and vfs:
                vfs.unlink(cleanup)
            else:
                os.unlink(cleanup)

def decompressor(fh, alg):
    if alg == 'UN':
        return fh
    elif alg == 'GZ':
        def generator(f):
            zd = zlib.decompressobj()
            for chunk in util.filechunkiter(f):
                yield zd.decompress(chunk)
    elif alg == 'BZ':
        def generator(f):
            zd = bz2.BZ2Decompressor()
            zd.decompress("BZ")
            for chunk in util.filechunkiter(f, 4096):
                yield zd.decompress(chunk)
    else:
        raise util.Abort("unknown bundle compression '%s'" % alg)
    return util.chunkbuffer(generator(fh))

class unbundle10(object):
    deltaheader = _BUNDLE10_DELTA_HEADER
    deltaheadersize = struct.calcsize(deltaheader)
    def __init__(self, fh, alg):
        self._stream = decompressor(fh, alg)
        self._type = alg
        self.callback = None
    def compressed(self):
        return self._type != 'UN'
    def read(self, l):
        return self._stream.read(l)
    def seek(self, pos):
        return self._stream.seek(pos)
    def tell(self):
        return self._stream.tell()
    def close(self):
        return self._stream.close()

    def chunklength(self):
        d = readexactly(self._stream, 4)
        l = struct.unpack(">l", d)[0]
        if l <= 4:
            if l:
                raise util.Abort(_("invalid chunk length %d") % l)
            return 0
        if self.callback:
            self.callback()
        return l - 4

    def changelogheader(self):
        """v10 does not have a changelog header chunk"""
        return {}

    def manifestheader(self):
        """v10 does not have a manifest header chunk"""
        return {}

    def filelogheader(self):
        """return the header of the filelogs chunk, v10 only has the filename"""
        l = self.chunklength()
        if not l:
            return {}
        fname = readexactly(self._stream, l)
        return {'filename': fname}

    def _deltaheader(self, headertuple, prevnode):
        node, p1, p2, cs = headertuple
        if prevnode is None:
            deltabase = p1
        else:
            deltabase = prevnode
        return node, p1, p2, deltabase, cs

    def deltachunk(self, prevnode):
        l = self.chunklength()
        if not l:
            return {}
        headerdata = readexactly(self._stream, self.deltaheadersize)
        header = struct.unpack(self.deltaheader, headerdata)
        delta = readexactly(self._stream, l - self.deltaheadersize)
        node, p1, p2, deltabase, cs = self._deltaheader(header, prevnode)
        return {'node': node, 'p1': p1, 'p2': p2, 'cs': cs,
                'deltabase': deltabase, 'delta': delta}

    def getchunks(self):
        """returns all the chunks contains in the bundle

        Used when you need to forward the binary stream to a file or another
        network API. To do so, it parse the changegroup data, otherwise it will
        block in case of sshrepo because it don't know the end of the stream.
        """
        # an empty chunkgroup is the end of the changegroup
        # a changegroup has at least 2 chunkgroups (changelog and manifest).
        # after that, an empty chunkgroup is the end of the changegroup
        empty = False
        count = 0
        while not empty or count <= 2:
            empty = True
            count += 1
            while True:
                chunk = getchunk(self)
                if not chunk:
                    break
                empty = False
                yield chunkheader(len(chunk))
                pos = 0
                while pos < len(chunk):
                    next = pos + 2**20
                    yield chunk[pos:next]
                    pos = next
            yield closechunk()

class headerlessfixup(object):
    def __init__(self, fh, h):
        self._h = h
        self._fh = fh
    def read(self, n):
        if self._h:
            d, self._h = self._h[:n], self._h[n:]
            if len(d) < n:
                d += readexactly(self._fh, n - len(d))
            return d
        return readexactly(self._fh, n)

def readbundle(fh, fname, vfs=None):
    header = readexactly(fh, 6)

    if not fname:
        fname = "stream"
        if not header.startswith('HG') and header.startswith('\0'):
            fh = headerlessfixup(fh, header)
            header = "HG10UN"
    elif vfs:
        fname = vfs.join(fname)

    magic, version, alg = header[0:2], header[2:4], header[4:6]

    if magic != 'HG':
        raise util.Abort(_('%s: not a Mercurial bundle') % fname)
    if version != '10':
        raise util.Abort(_('%s: unknown bundle version %s') % (fname, version))
    return unbundle10(fh, alg)

class bundle10(object):
    deltaheader = _BUNDLE10_DELTA_HEADER
    def __init__(self, repo, bundlecaps=None):
        """Given a source repo, construct a bundler.

        bundlecaps is optional and can be used to specify the set of
        capabilities which can be used to build the bundle.
        """
        # Set of capabilities we can use to build the bundle.
        if bundlecaps is None:
            bundlecaps = set()
        self._bundlecaps = bundlecaps
        self._changelog = repo.changelog
        self._manifest = repo.manifest
        reorder = repo.ui.config('bundle', 'reorder', 'auto')
        if reorder == 'auto':
            reorder = None
        else:
            reorder = util.parsebool(reorder)
        self._repo = repo
        self._reorder = reorder
        self._progress = repo.ui.progress
    def close(self):
        return closechunk()

    def fileheader(self, fname):
        return chunkheader(len(fname)) + fname

    def group(self, nodelist, revlog, lookup, units=None, reorder=None):
        """Calculate a delta group, yielding a sequence of changegroup chunks
        (strings).

        Given a list of changeset revs, return a set of deltas and
        metadata corresponding to nodes. The first delta is
        first parent(nodelist[0]) -> nodelist[0], the receiver is
        guaranteed to have this parent as it has all history before
        these changesets. In the case firstparent is nullrev the
        changegroup starts with a full revision.

        If units is not None, progress detail will be generated, units specifies
        the type of revlog that is touched (changelog, manifest, etc.).
        """
        # if we don't have any revisions touched by these changesets, bail
        if len(nodelist) == 0:
            yield self.close()
            return

        # for generaldelta revlogs, we linearize the revs; this will both be
        # much quicker and generate a much smaller bundle
        if (revlog._generaldelta and reorder is not False) or reorder:
            dag = dagutil.revlogdag(revlog)
            revs = set(revlog.rev(n) for n in nodelist)
            revs = dag.linearize(revs)
        else:
            revs = sorted([revlog.rev(n) for n in nodelist])

        # add the parent of the first rev
        p = revlog.parentrevs(revs[0])[0]
        revs.insert(0, p)

        # build deltas
        total = len(revs) - 1
        msgbundling = _('bundling')
        for r in xrange(len(revs) - 1):
            if units is not None:
                self._progress(msgbundling, r + 1, unit=units, total=total)
            prev, curr = revs[r], revs[r + 1]
            linknode = lookup(revlog.node(curr))
            for c in self.revchunk(revlog, curr, prev, linknode):
                yield c

        yield self.close()

    # filter any nodes that claim to be part of the known set
    def prune(self, revlog, missing, commonrevs, source):
        rr, rl = revlog.rev, revlog.linkrev
        return [n for n in missing if rl(rr(n)) not in commonrevs]

    def generate(self, commonrevs, clnodes, fastpathlinkrev, source):
        '''yield a sequence of changegroup chunks (strings)'''
        repo = self._repo
        cl = self._changelog
        mf = self._manifest
        reorder = self._reorder
        progress = self._progress

        # for progress output
        msgbundling = _('bundling')

        mfs = {} # needed manifests
        fnodes = {} # needed file nodes
        changedfiles = set()

        # Callback for the changelog, used to collect changed files and manifest
        # nodes.
        # Returns the linkrev node (identity in the changelog case).
        def lookupcl(x):
            c = cl.read(x)
            changedfiles.update(c[3])
            # record the first changeset introducing this manifest version
            mfs.setdefault(c[0], x)
            return x

        # Callback for the manifest, used to collect linkrevs for filelog
        # revisions.
        # Returns the linkrev node (collected in lookupcl).
        def lookupmf(x):
            clnode = mfs[x]
            if not fastpathlinkrev:
                mdata = mf.readfast(x)
                for f, n in mdata.iteritems():
                    if f in changedfiles:
                        # record the first changeset introducing this filelog
                        # version
                        fnodes[f].setdefault(n, clnode)
            return clnode

        for chunk in self.group(clnodes, cl, lookupcl, units=_('changesets'),
                                reorder=reorder):
            yield chunk
        progress(msgbundling, None)

        for f in changedfiles:
            fnodes[f] = {}
        mfnodes = self.prune(mf, mfs, commonrevs, source)
        for chunk in self.group(mfnodes, mf, lookupmf, units=_('manifests'),
                                reorder=reorder):
            yield chunk
        progress(msgbundling, None)

        mfs.clear()
        needed = set(cl.rev(x) for x in clnodes)

        def linknodes(filerevlog, fname):
            if fastpathlinkrev:
                llr = filerevlog.linkrev
                def genfilenodes():
                    for r in filerevlog:
                        linkrev = llr(r)
                        if linkrev in needed:
                            yield filerevlog.node(r), cl.node(linkrev)
                fnodes[fname] = dict(genfilenodes())
            return fnodes.get(fname, {})

        for chunk in self.generatefiles(changedfiles, linknodes, commonrevs,
                                        source):
            yield chunk

        yield self.close()
        progress(msgbundling, None)

        if clnodes:
            repo.hook('outgoing', node=hex(clnodes[0]), source=source)

    def generatefiles(self, changedfiles, linknodes, commonrevs, source):
        repo = self._repo
        progress = self._progress
        reorder = self._reorder
        msgbundling = _('bundling')

        total = len(changedfiles)
        # for progress output
        msgfiles = _('files')
        for i, fname in enumerate(sorted(changedfiles)):
            filerevlog = repo.file(fname)
            if not filerevlog:
                raise util.Abort(_("empty or missing revlog for %s") % fname)

            linkrevnodes = linknodes(filerevlog, fname)
            # Lookup for filenodes, we collected the linkrev nodes above in the
            # fastpath case and with lookupmf in the slowpath case.
            def lookupfilelog(x):
                return linkrevnodes[x]

            filenodes = self.prune(filerevlog, linkrevnodes, commonrevs, source)
            if filenodes:
                progress(msgbundling, i + 1, item=fname, unit=msgfiles,
                         total=total)
                yield self.fileheader(fname)
                for chunk in self.group(filenodes, filerevlog, lookupfilelog,
                                        reorder=reorder):
                    yield chunk

    def revchunk(self, revlog, rev, prev, linknode):
        node = revlog.node(rev)
        p1, p2 = revlog.parentrevs(rev)
        base = prev

        prefix = ''
        if base == nullrev:
            delta = revlog.revision(node)
            prefix = mdiff.trivialdiffheader(len(delta))
        else:
            delta = revlog.revdiff(base, rev)
        p1n, p2n = revlog.parents(node)
        basenode = revlog.node(base)
        meta = self.builddeltaheader(node, p1n, p2n, basenode, linknode)
        meta += prefix
        l = len(meta) + len(delta)
        yield chunkheader(l)
        yield meta
        yield delta
    def builddeltaheader(self, node, p1n, p2n, basenode, linknode):
        # do nothing with basenode, it is implicitly the previous one in HG10
        return struct.pack(self.deltaheader, node, p1n, p2n, linknode)

def _changegroupinfo(repo, nodes, source):
    if repo.ui.verbose or source == 'bundle':
        repo.ui.status(_("%d changesets found\n") % len(nodes))
    if repo.ui.debugflag:
        repo.ui.debug("list of changesets:\n")
        for node in nodes:
            repo.ui.debug("%s\n" % hex(node))

def getsubset(repo, outgoing, bundler, source, fastpath=False):
    repo = repo.unfiltered()
    commonrevs = outgoing.common
    csets = outgoing.missing
    heads = outgoing.missingheads
    # We go through the fast path if we get told to, or if all (unfiltered
    # heads have been requested (since we then know there all linkrevs will
    # be pulled by the client).
    heads.sort()
    fastpathlinkrev = fastpath or (
            repo.filtername is None and heads == sorted(repo.heads()))

    repo.hook('preoutgoing', throw=True, source=source)
    _changegroupinfo(repo, csets, source)
    gengroup = bundler.generate(commonrevs, csets, fastpathlinkrev, source)
    return unbundle10(util.chunkbuffer(gengroup), 'UN')

def changegroupsubset(repo, roots, heads, source):
    """Compute a changegroup consisting of all the nodes that are
    descendants of any of the roots and ancestors of any of the heads.
    Return a chunkbuffer object whose read() method will return
    successive changegroup chunks.

    It is fairly complex as determining which filenodes and which
    manifest nodes need to be included for the changeset to be complete
    is non-trivial.

    Another wrinkle is doing the reverse, figuring out which changeset in
    the changegroup a particular filenode or manifestnode belongs to.
    """
    cl = repo.changelog
    if not roots:
        roots = [nullid]
    # TODO: remove call to nodesbetween.
    csets, roots, heads = cl.nodesbetween(roots, heads)
    discbases = []
    for n in roots:
        discbases.extend([p for p in cl.parents(n) if p != nullid])
    outgoing = discovery.outgoing(cl, discbases, heads)
    bundler = bundle10(repo)
    return getsubset(repo, outgoing, bundler, source)

def getlocalbundle(repo, source, outgoing, bundlecaps=None):
    """Like getbundle, but taking a discovery.outgoing as an argument.

    This is only implemented for local repos and reuses potentially
    precomputed sets in outgoing."""
    if not outgoing.missing:
        return None
    bundler = bundle10(repo, bundlecaps)
    return getsubset(repo, outgoing, bundler, source)

def getbundle(repo, source, heads=None, common=None, bundlecaps=None):
    """Like changegroupsubset, but returns the set difference between the
    ancestors of heads and the ancestors common.

    If heads is None, use the local heads. If common is None, use [nullid].

    The nodes in common might not all be known locally due to the way the
    current discovery protocol works.
    """
    cl = repo.changelog
    if common:
        hasnode = cl.hasnode
        common = [n for n in common if hasnode(n)]
    else:
        common = [nullid]
    if not heads:
        heads = cl.heads()
    outgoing = discovery.outgoing(cl, common, heads)
    return getlocalbundle(repo, source, outgoing, bundlecaps=bundlecaps)

def changegroup(repo, basenodes, source):
    # to avoid a race we use changegroupsubset() (issue1320)
    return changegroupsubset(repo, basenodes, repo.heads(), source)

def addchangegroupfiles(repo, source, revmap, trp, pr, needfiles):
    revisions = 0
    files = 0
    while True:
        chunkdata = source.filelogheader()
        if not chunkdata:
            break
        f = chunkdata["filename"]
        repo.ui.debug("adding %s revisions\n" % f)
        pr()
        fl = repo.file(f)
        o = len(fl)
        if not fl.addgroup(source, revmap, trp):
            raise util.Abort(_("received file revlog group is empty"))
        revisions += len(fl) - o
        files += 1
        if f in needfiles:
            needs = needfiles[f]
            for new in xrange(o, len(fl)):
                n = fl.node(new)
                if n in needs:
                    needs.remove(n)
                else:
                    raise util.Abort(
                        _("received spurious file revlog entry"))
            if not needs:
                del needfiles[f]
    repo.ui.progress(_('files'), None)

    for f, needs in needfiles.iteritems():
        fl = repo.file(f)
        for n in needs:
            try:
                fl.rev(n)
            except error.LookupError:
                raise util.Abort(
                    _('missing file data for %s:%s - run hg verify') %
                    (f, hex(n)))

    return revisions, files

def addchangegroup(repo, source, srctype, url, emptyok=False):
    """Add the changegroup returned by source.read() to this repo.
    srctype is a string like 'push', 'pull', or 'unbundle'.  url is
    the URL of the repo where this changegroup is coming from.

    Return an integer summarizing the change to this repo:
    - nothing changed or no source: 0
    - more heads than before: 1+added heads (2..n)
    - fewer heads than before: -1-removed heads (-2..-n)
    - number of heads stays the same: 1
    """
    repo = repo.unfiltered()
    def csmap(x):
        repo.ui.debug("add changeset %s\n" % short(x))
        return len(cl)

    def revmap(x):
        return cl.rev(x)

    if not source:
        return 0

    repo.hook('prechangegroup', throw=True, source=srctype, url=url)

    changesets = files = revisions = 0
    efiles = set()

    # write changelog data to temp files so concurrent readers will not see
    # inconsistent view
    cl = repo.changelog
    cl.delayupdate()
    oldheads = cl.heads()

    tr = repo.transaction("\n".join([srctype, util.hidepassword(url)]))
    try:
        trp = weakref.proxy(tr)
        # pull off the changeset group
        repo.ui.status(_("adding changesets\n"))
        clstart = len(cl)
        class prog(object):
            step = _('changesets')
            count = 1
            ui = repo.ui
            total = None
            def __call__(repo):
                repo.ui.progress(repo.step, repo.count, unit=_('chunks'),
                                 total=repo.total)
                repo.count += 1
        pr = prog()
        source.callback = pr

        source.changelogheader()
        srccontent = cl.addgroup(source, csmap, trp)
        if not (srccontent or emptyok):
            raise util.Abort(_("received changelog group is empty"))
        clend = len(cl)
        changesets = clend - clstart
        for c in xrange(clstart, clend):
            efiles.update(repo[c].files())
        efiles = len(efiles)
        repo.ui.progress(_('changesets'), None)

        # pull off the manifest group
        repo.ui.status(_("adding manifests\n"))
        pr.step = _('manifests')
        pr.count = 1
        pr.total = changesets # manifests <= changesets
        # no need to check for empty manifest group here:
        # if the result of the merge of 1 and 2 is the same in 3 and 4,
        # no new manifest will be created and the manifest group will
        # be empty during the pull
        source.manifestheader()
        repo.manifest.addgroup(source, revmap, trp)
        repo.ui.progress(_('manifests'), None)

        needfiles = {}
        if repo.ui.configbool('server', 'validate', default=False):
            # validate incoming csets have their manifests
            for cset in xrange(clstart, clend):
                mfest = repo.changelog.read(repo.changelog.node(cset))[0]
                mfest = repo.manifest.readdelta(mfest)
                # store file nodes we must see
                for f, n in mfest.iteritems():
                    needfiles.setdefault(f, set()).add(n)

        # process the files
        repo.ui.status(_("adding file changes\n"))
        pr.step = _('files')
        pr.count = 1
        pr.total = efiles
        source.callback = None

        newrevs, newfiles = addchangegroupfiles(repo, source, revmap, trp, pr,
                                                needfiles)
        revisions += newrevs
        files += newfiles

        dh = 0
        if oldheads:
            heads = cl.heads()
            dh = len(heads) - len(oldheads)
            for h in heads:
                if h not in oldheads and repo[h].closesbranch():
                    dh -= 1
        htext = ""
        if dh:
            htext = _(" (%+d heads)") % dh

        repo.ui.status(_("added %d changesets"
                         " with %d changes to %d files%s\n")
                         % (changesets, revisions, files, htext))
        repo.invalidatevolatilesets()

        if changesets > 0:
            p = lambda: cl.writepending() and repo.root or ""
            repo.hook('pretxnchangegroup', throw=True,
                      node=hex(cl.node(clstart)), source=srctype,
                      url=url, pending=p)

        added = [cl.node(r) for r in xrange(clstart, clend)]
        publishing = repo.ui.configbool('phases', 'publish', True)
        if srctype in ('push', 'serve'):
            # Old servers can not push the boundary themselves.
            # New servers won't push the boundary if changeset already
            # exists locally as secret
            #
            # We should not use added here but the list of all change in
            # the bundle
            if publishing:
                phases.advanceboundary(repo, phases.public, srccontent)
            else:
                phases.advanceboundary(repo, phases.draft, srccontent)
                phases.retractboundary(repo, phases.draft, added)
        elif srctype != 'strip':
            # publishing only alter behavior during push
            #
            # strip should not touch boundary at all
            phases.retractboundary(repo, phases.draft, added)

        # make changelog see real files again
        cl.finalize(trp)

        tr.close()

        if changesets > 0:
            if srctype != 'strip':
                # During strip, branchcache is invalid but coming call to
                # `destroyed` will repair it.
                # In other case we can safely update cache on disk.
                branchmap.updatecache(repo.filtered('served'))
            def runhooks():
                # These hooks run when the lock releases, not when the
                # transaction closes. So it's possible for the changelog
                # to have changed since we last saw it.
                if clstart >= len(repo):
                    return

                # forcefully update the on-disk branch cache
                repo.ui.debug("updating the branch cache\n")
                repo.hook("changegroup", node=hex(cl.node(clstart)),
                          source=srctype, url=url)

                for n in added:
                    repo.hook("incoming", node=hex(n), source=srctype,
                              url=url)

                newheads = [h for h in repo.heads() if h not in oldheads]
                repo.ui.log("incoming",
                            "%s incoming changes - new heads: %s\n",
                            len(added),
                            ', '.join([hex(c[:6]) for c in newheads]))
            repo._afterlock(runhooks)

    finally:
        tr.release()
    # never return 0 here:
    if dh < 0:
        return dh - 1
    else:
        return dh + 1